Remanentia
Persistent memory and retrieval for local AI systems. Current project state tracks 2,062 tests, 14 Rust crates, vector search, LongMemEval R11 at 72.2% overall, and temporal retrieval at 65.4%.
Builder CV + GOTM ecosystem
Miroslav Šotek
Independent builder behind the God of the Math ecosystem: manuscripts, research engines, memory systems, factual-control software, stochastic-computing neural work, control dynamics, quantum-control experiments, audio entrainment, public websites, and local infrastructure. Remanentia is one node in that portfolio, not a separate side project.
Profile
Research, systems engineering, and product work under one roof.
GOTM is the working archive and build environment behind ANULUM. It contains the theory, code, public sites, manuscripts, benchmark records, deployment mirrors, server work, product pages, and research traces needed to move ideas from private notes into systems that can be tested. The role is deliberately mixed: design the theory, write the code, run the systems, document the limits, and keep the public boundary honest.
1996-2026concept development horizon
2020-2026active code authorship horizon
GOTMsingle ecosystem, many surfaces
CH / LIANULUM operating base
Portfolio map
The entire GOTM ecosystem is the portfolio.
The complexity is intentional. Memory software, factual verification, stochastic neural computation, control research, manuscripts, civic publication, public websites, and server infrastructure all support the same larger work. Remanentia cannot stand alone if model output is unchecked. Factual control is weaker without memory. Control systems need simulation and measurement. Research products need public pages, licences, support paths, and operational infrastructure.
Director Class AI
A verification layer for generated answers and RAG pipelines. It checks whether an answer is supported by evidence before the output is accepted for operational or public use.
SC-NeuroCore
Stochastic-computing neural infrastructure, published as `sc-neurocore-engine`, used as the computational base for efficient neural and hardware-facing experiments.
SCPN control family
Fusion-core, phase orchestration, classical control, infinity control, and quantum-control work sit around a shared need: model coupled dynamics, test interventions, and keep assumptions visible.
Fluctara
Audio entrainment with a Rust engine and production API work. Its need is different from Remanentia, but the engineering question is similar: convert theory into a measured, user-facing system.
God of the Math corpus
The research and manuscript body behind the code. It holds the theoretical continuity: SCPN, coherence-carrier work, control ideas, experimental notes, and long-term concept development.
Webmaster layer
Product pages, legal pages, docs entrances, pricing, contact routes, and public mirrors. This layer turns private research and local code into something reviewers and users can inspect.
Local servers and deployment
Workstations, local inference, ML350 server work, backup mirrors, FTP deployment, benchmark machines, and operational records. The portfolio depends on running systems, not only repositories.
Parazit.sk and civic work
A separate civic publication track inside the wider portfolio. It exists because technical systems are not enough; public records, naming, and accessible presentation also matter.
Why it is needed
High-stakes AI cannot be only a model call.
When a system advises a person, answers a customer, summarises research, routes an operational task, or touches private records, fluency is not enough. The missing pieces are memory, source selection, verification, policy, latency measurement, and a way to inspect what happened after the fact.
That is the GOTM portfolio thesis: build the pieces that make AI systems less dependent on luck. Remanentia gives memory. Director Class AI gives factual control. The SCPN and SC-NeuroCore work explores alternative computation and control. The manuscript corpus keeps the long arc coherent. The public sites make the work legible enough for users, reviewers, partners, and funders.
1Remember the right source material instead of relying on a single prompt.
2Verify generated text against evidence before using it.
3Measure retrieval, latency, error modes, and deployment constraints.
4Turn the GOTM research ecosystem into services that people can evaluate and fund.
Functions
Expected functions across the portfolio.
Memory systems
Indexing, retrieval, vector search, compiled facts, operational recall, and public/private corpus boundaries.
Verification systems
Evidence checks, factual-consistency scoring, answer gating, refusal paths, and commercial licensing for closed deployments.
Compute systems
Rust acceleration, stochastic-computing models, WebGPU-facing work, and hardware-aware benchmarking.
Control systems
Phase dynamics, feedback loops, quantum-control experiments, and simulation surfaces for coupled systems.
Product systems
Documentation, pricing, support paths, deployment pages, legal pages, payment channels, and public project positioning.
Research systems
Benchmarks, reproducibility notes, manuscript work, peer-review channels, and long-term concept continuity.
Portfolio systems
Cross-project continuity, public mirrors, backups, session records, shared context, and the operational discipline needed to keep the whole ecosystem coherent.
Impact and potential
The near-term value is practical; the long-term value is structural.
For companiesSafer AI answers over private knowledge, clearer data boundaries, and deployment evidence that can be reviewed.
For researchersMemory benchmarks, factual-control methods, and control-system experiments that can be compared instead of only described.
For local operatorsSelf-hosted AI infrastructure that works with real constraints: hardware limits, corpus size, latency, privacy, and maintenance.
For the GOTM portfolioA single body of work whose pieces can support each other: memory improves verification, verification makes deployment safer, manuscripts preserve theory, and deployment reveals what research still has to solve.
Collaboration fit
Useful conversations are concrete: a memory failure, a retrieval problem, a factual-control requirement, a benchmark to reproduce, a hardware constraint, a research review, or a deployment that cannot afford unsupported answers.